Using social connections to improve collaborative filtering
Kanish Manuja, Arnab Bhattacharya · 2015
In this paper, we test the hypothesis that for a particular item recommendation, it matters more how a friend (i.e., another user who is socially connected) has rated than a random user. To test this, we propose a matrix factorization based collaborative filtering approach that utilizes the social connections as an additional term in the objective function that aims to minimize the difference between predicted and actual rankings. We obtain an improvement of 3.09% in accuracy for the Epinions dataset which has trust-based social connections and 0.61% on the Douban dataset which has friendship-based social connections.